Re: [PATCH 13/22] clk: mediatek: Add MT8195 nnasys clock support - Chen-Yu Tsai

From: Chen-Yu Tsai <wenst@chromium.org>
To: Chun-Jie Chen <chun-jie.chen@mediatek.com>
Cc: Matthias Brugger <matthias.bgg@gmail.com>,
	Stephen Boyd <sboyd@kernel.org>,
	Nicolas Boichat <drinkcat@chromium.org>,
	Rob Herring <robh+dt@kernel.org>,
	linux-arm-kernel@lists.infradead.org,
	LKML <linux-kernel@vger.kernel.org>,
	linux-mediatek@lists.infradead.org, linux-clk@vger.kernel.org,
	devicetree@vger.kernel.org, srv_heupstream@mediatek.com,
	Project_Global_Chrome_Upstream_Group@mediatek.com
Subject: Re: [PATCH 13/22] clk: mediatek: Add MT8195 nnasys clock support
Date: Fri, 9 Jul 2021 16:24:12 +0800	[thread overview]
Message-ID: <CAGXv+5GR7gRnAMJSJTPbq1QUL58Fq5oejuAvaEwLHTYNMwVBzQ@mail.gmail.com> (raw)
In-Reply-To: <20210616224743.5109-14-chun-jie.chen@mediatek.com>

Hi,

On Thu, Jun 17, 2021 at 7:00 AM Chun-Jie Chen
<chun-jie.chen@mediatek.com> wrote:
>
> Add MT8195 nnasys clock provider
>
> Signed-off-by: Chun-Jie Chen <chun-jie.chen@mediatek.com>
> ---
>  drivers/clk/mediatek/Kconfig          |   6 ++
>  drivers/clk/mediatek/Makefile         |   1 +
>  drivers/clk/mediatek/clk-mt8195-nna.c | 128 ++++++++++++++++++++++++++
>  3 files changed, 135 insertions(+)
>  create mode 100644 drivers/clk/mediatek/clk-mt8195-nna.c
>
> diff --git a/drivers/clk/mediatek/Kconfig b/drivers/clk/mediatek/Kconfig
> index 7cb745d47770..d34517728f4a 100644
> --- a/drivers/clk/mediatek/Kconfig
> +++ b/drivers/clk/mediatek/Kconfig
> @@ -636,6 +636,12 @@ config COMMON_CLK_MT8195_SCP_ADSP
>         help
>           This driver supports MediaTek MT8195 scp_adsp clocks.
>
> +config COMMON_CLK_MT8195_NNASYS
> +       bool "Clock driver for MediaTek MT8195 nnasys"
> +       depends on COMMON_CLK_MT8195
> +       help
> +         This driver supports MediaTek MT8195 nnasys clocks.
> +
>  config COMMON_CLK_MT8516
>         bool "Clock driver for MediaTek MT8516"
>         depends on ARCH_MEDIATEK || COMPILE_TEST

Same comments about commit log and Kconfig option apply.

> diff --git a/drivers/clk/mediatek/Makefile b/drivers/clk/mediatek/Makefile
> index 058ff55468a2..49e585a7ac8e 100644
> --- a/drivers/clk/mediatek/Makefile
> +++ b/drivers/clk/mediatek/Makefile
> @@ -89,5 +89,6 @@ obj-$(CONFIG_COMMON_CLK_MT8195_IMGSYS) += clk-mt8195-img.o
>  obj-$(CONFIG_COMMON_CLK_MT8195_IPESYS) += clk-mt8195-ipe.o
>  obj-$(CONFIG_COMMON_CLK_MT8195_MFGCFG) += clk-mt8195-mfg.o
>  obj-$(CONFIG_COMMON_CLK_MT8195_SCP_ADSP) += clk-mt8195-scp_adsp.o
> +obj-$(CONFIG_COMMON_CLK_MT8195_NNASYS) += clk-mt8195-nna.o
>  obj-$(CONFIG_COMMON_CLK_MT8516) += clk-mt8516.o
>  obj-$(CONFIG_COMMON_CLK_MT8516_AUDSYS) += clk-mt8516-aud.o
> diff --git a/drivers/clk/mediatek/clk-mt8195-nna.c b/drivers/clk/mediatek/clk-mt8195-nna.c
> new file mode 100644
> index 000000000000..4210c6cf5ef4
> --- /dev/null
> +++ b/drivers/clk/mediatek/clk-mt8195-nna.c
> @@ -0,0 +1,128 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +//
> +// Copyright (c) 2021 MediaTek Inc.
> +// Author: Chun-Jie Chen <chun-jie.chen@mediatek.com>
> +
> +#include <linux/clk-provider.h>
> +#include <linux/platform_device.h>
> +
> +#include "clk-mtk.h"
> +#include "clk-gate.h"
> +
> +#include <dt-bindings/clock/mt8195-clk.h>
> +
> +static const struct mtk_gate_regs nna0_cg_regs = {
> +       .set_ofs = 0x104,
> +       .clr_ofs = 0x104,
> +       .sta_ofs = 0x104,

You are using the _no_setclr_ops variant. That means the .set_ofs and
.clr_ofs fields don't have any meaning and are not used. Please just
don't initialize them.

I think this applies to some of the other patches as well.

> +};
> +
> +static const struct mtk_gate_regs nna1_cg_regs = {
> +       .set_ofs = 0x110,
> +       .clr_ofs = 0x110,
> +       .sta_ofs = 0x110,
> +};
> +
> +static const struct mtk_gate_regs nna2_cg_regs = {
> +       .set_ofs = 0x90,
> +       .clr_ofs = 0x90,
> +       .sta_ofs = 0x90,
> +};
> +
> +static const struct mtk_gate_regs nna3_cg_regs = {
> +       .set_ofs = 0x94,
> +       .clr_ofs = 0x94,
> +       .sta_ofs = 0x94,
> +};
> +
> +static const struct mtk_gate_regs nna4_cg_regs = {
> +       .set_ofs = 0x98,
> +       .clr_ofs = 0x98,
> +       .sta_ofs = 0x98,
> +};
> +
> +static const struct mtk_gate_regs nna5_cg_regs = {
> +       .set_ofs = 0x9c,
> +       .clr_ofs = 0x9c,
> +       .sta_ofs = 0x9c,
> +};
> +
> +static const struct mtk_gate_regs nna6_cg_regs = {
> +       .set_ofs = 0xa0,
> +       .clr_ofs = 0xa0,
> +       .sta_ofs = 0xa0,
> +};
> +
> +static const struct mtk_gate_regs nna7_cg_regs = {
> +       .set_ofs = 0xa4,
> +       .clr_ofs = 0xa4,
> +       .sta_ofs = 0xa4,
> +};

Unfortunately this hardware block is not documented in the datasheets,
so I can't verify these register offsets.

> +
> +#define GATE_NNA0(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna0_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA1(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna1_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA2(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna2_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA3(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna3_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA4(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna4_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA5(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna5_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA6(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna6_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)
> +
> +#define GATE_NNA7(_id, _name, _parent, _shift)                 \
> +       GATE_MTK(_id, _name, _parent, &nna7_cg_regs, _shift, &mtk_clk_gate_ops_no_setclr_inv)

Defining a bunch of macros, only to have each of them used once doesn't
provide any gains with regards to reusing code. Please consider just using
GATE_MTK directly below.

In the general case I would probably make some derivatives of GATE_MTK,
such that the ops field (which seems to be the longest) is hidden, but
all the other fields are exposed.

> +static const struct mtk_gate nna_clks[] = {
> +       /* NNA0 */
> +       GATE_NNA0(CLK_NNA_F26M, "nna_f26m", "clk26m", 0),
> +       /* NNA1 */
> +       GATE_NNA1(CLK_NNA_AXI, "nna_axi", "axi_sel", 0),
> +       /* NNA2 */
> +       GATE_NNA2(CLK_NNA_NNA0, "nna_nna0", "nna0_sel", 0),
> +       /* NNA3 */
> +       GATE_NNA3(CLK_NNA_NNA1, "nna_nna1", "nna0_sel", 0),

Is this correct? I would expect the clock for nna1 to be fed from
"nna1_sel".

ChenYu

> +       /* NNA4 */
> +       GATE_NNA4(CLK_NNA_NNA0_EMI, "nna_nna0_emi", "mem_466m", 0),
> +       GATE_NNA4(CLK_NNA_CKGEN_MEM, "nna_ckgen_mem", "mem_466m", 4),
> +       /* NNA5 */
> +       GATE_NNA5(CLK_NNA_NNA1_EMI, "nna_nna1_emi", "mem_466m", 0),
> +       /* NNA6 */
> +       GATE_NNA6(CLK_NNA_NNA0_AXI, "nna_nna0_axi", "axi_sel", 0),
> +       /* NNA7 */
> +       GATE_NNA7(CLK_NNA_NNA1_AXI, "nna_nna1_axi", "axi_sel", 0),
> +};
> +
> +static const struct mtk_clk_desc nna_desc = {
> +       .clks = nna_clks,
> +       .num_clks = ARRAY_SIZE(nna_clks),
> +};
> +
> +static const struct of_device_id of_match_clk_mt8195_nna[] = {
> +       {
> +               .compatible = "mediatek,mt8195-nnasys",
> +               .data = &nna_desc,
> +       }, {
> +               /* sentinel */
> +       }
> +};
> +
> +static struct platform_driver clk_mt8195_nna_drv = {
> +       .probe = mtk_clk_simple_probe,
> +       .driver = {
> +               .name = "clk-mt8195-nna",
> +               .of_match_table = of_match_clk_mt8195_nna,
> +       },
> +};
> +
> +builtin_platform_driver(clk_mt8195_nna_drv);
> --
> 2.18.0
> _______________________________________________
> Linux-mediatek mailing list
> Linux-mediatek@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-mediatek